CSE 5324: Software Engineering I. (Analysis, Design, Creation)

Transcription

1 CSE 5324: Software Engineering I (Analysis, Design, Creation)

2 Process and process improvement

3 The way you get better is practice, practice, practice. (And learn from your mistakes) Defining your process: what steps are done when? what is the result? Learn from (your) history What can you measure? Transactions/sec. Average module (or method) size (or complexity) Time/module (or method) Reuse rate Time per phase Fault density Requirements "creep" (changes in new/changed function points) Schedule slips and much more Quantitative measurements What can change? What could be improved? Variability, quality, etc. Predictive: what can one predict and plan knowing history

4 Many tools are statistical: based on broad sample of industry, many types of projects, many types of programming languages, styles, methods, etc. COCOMO FP counting... Your history is much more accurate Set targets Set baselines Goals

5 CSE 5324: Software Engineering I (Analysis, Design, Creation)

6 Review: SQA and Testing Cleanroom Formal Methods Real Time Maintenance Process and process improvement Exam 3: November 18 Primarily Text: Chap 8, 9.2, 10, , 4.5 (Z) Real Time, Z, Test, Maintenance, Process measurement and improvement Review

7 Why Grades are not important (But what is...)

8 Real-Time Systems: What are they? When (where) are they used? How are they different? (From non-rt)

9 Many (most) real-time systems are embedded Where are they seen: where real-world events are time critical: navigation (air, space) medical telecomm, network games Often classified into: Soft (RT) Hard (RT) Usually both (and non-rt requirements also) Degraded operation versus Incorrect results

10 Often viewed as stimulus - response system Stimuli: Periodic (predictable time intervals) Aperiodic (irregular) Response: Actuator Classification: Monitoring systems Data Acquisition Command and Control etc.

11 Usually hardware and software done in parallel (and tradeoffs are made) Establish system requirements Partition Requirements Software Reqs Hardware Reqs Software Design Hardware Design Steps: Identify Stimuli (to process) and associated response For each stimuli and response identify timing constraints Group stimuli and response into concurrent processes For each, design algorithms (usually early in design process, may make tradeoffs) Design scheduling system to ensure processes started on time Integrate with a real-time executive (OS)

12 Implementation overhead is critical: Design tools and methodolgies Programming language: Is C appropriate? (NO) C++? (Even worse) Ada Ada 83 was not (design errors) but Ada 95 is What is needed: HLL synchronization OS interface constructs: mutual exclusion, IPC, start process by time, scheduler changes, etc. Interrupt handling

13 OS: (Not Unix!) Real time clock Interrupt handler Scheduler Resource manager Dispatcher May include: Configuration manger Static or dynamic Fault manager Hardware and software

14 Some Issues: Is OO a good idea for RT? Timing: how to make estimates: Counting instructions (pipelines, etc.) Compiler helping Analysis Simulate Use good tools:

15 CSE 5324: Software Engineering I (Analysis, Design, Creation)

16 News (you can use) "Last week NASA announced that the "human" error stemming from space engineers using two sets of measurements -- one utilizing miles and the other kilometers -- caused the loss of the Mars Climate Orbiter spacecraft last week, NASA said Thursday." The teams, located at the National Aeronautics and Space Administration's Jet Propulsion Laboratory in Pasadena and at Lockheed Martin Astronautics in Colorado, complicated matters further by failing to realize the error, the agency said in a statement. The $125 million orbiter, intended to serve as the first interplanetary weather satellite, is believed to have broken up when it hit the Martian atmosphere last week after an approach that was too near the surface.

17 Today's question is: What kind of error was this? How could/(should) it have been prevented?

18 Object Oriented Methodolgy: Terms, ideas, techniques OO Concepts, OO Analysis, OO Design Design Concepts, Design Methods (Structured) Real Time Analysis and Design

19 What is designed: Data Architecture Interface Procedural

20 Object Oriented Design: Find objects, factor into good classes, define class hierarchies and interfaces, make it reusable, modifiable. OO Design architecture stresses object relations rather than flow of control What is different: (Fichman, Kemerer) (From SA) Module Hierarchy representation Data Definition Spec Procedural Spec End-to-end processing sequences Class and Hierarchy definition Operations assigned to classes

21 Bertrand Meyer suggests: Decomposable design method makes easier subproblems Composable reusable Understandable by module Continuity coupling (local) Protection coupling (isolation) To achieve: Few, small, explicit interfaces. Information hiding. OOD: Booch Coad Yourdon Rumbaugh Jackson design a part of analysis! Coad suggest (for any method): Define: Problem Domain User I/F Task Management Data Management

22 Design Patterns: (This is another of the SE Buzz words!) Recurring and reusing of classes and objects (Gamma) Name of pattern Problem to which applied Characteristics Consequences of applying Should one use inheritance or composition? (When both possible) Use what is general, or make it specific?

23 Compare SA, SD versus OO: Hospital system: Patient record system. Spreadsheet

24 Is OO Important? OO Prog languages: JAVA, et al OO Interfaces: Windowing systems CORBA Distributed objects

25 Cleanroom: (Formal Methods) Requirements: Remember: What to do. (what, NOT how.) What is wrong with what is done now? Vague Ambiguity Incomplete Contradictions Mixed Abstraction and methods Clients must agree to requirements, Developers must be able to design from them. Word descriptions (suffer from above) Structured Analysis O-O Analysis

26 Cleanroom Software Engineering Build correct software as it is developed. Use Mathematical Notation Verify Correctness (Statistical Quality Certification) Formal Design and verification Code Generation, Inspection, Verification Statistical Test Planning, Usage Testing Certification, Integration

27 Techniques used in Cleanroom: Incremental life-cycle Formal Techniques for Specification and Design Non-Execution Based Module-Testing (Code reading, walkthroughs, inspections, proofs) Module not compiled until AFTER passed inspection

28 Does Cleanroom work? US Naval Underwater systems Center: 18 Faults detected when design functionally verified (Informal proofs used mostly, only when participants unsure of correctness were full mathematical proofs used) 19 more faults during walkthrough (about 2KLOC) NO compilation errors NO Failures during execution Does it scale? Testing Fault Rate (Faults detected per KLOC) This is different between Cleanroom and other Development techniques. Ericsson Telecom OS32 350,000 Lines of code 18 months, team of 70 people Testing Fault Rate 1.0 Faults per KLOC (Navy was 0.0) Linger reports average Testing Fault Rate 2.3 Faults per KLOC based on 17 products of 1 million LOC. (This is very high quality)

29 What s not so good about Cleanroom: Problems?: Henderson: Too theoretical, mathematical, radical for real-world No Unit testing for developers Maturity of software developers, industry

30 Strategy: Increment Planning example Requirements Gathering Box structure Specification Formal Design Correctness Verification Code Generation, Inspect, Verify Statistical Test Planning Statistical Usage Testing Certification, Integration

31 Strategy: Increment Planning example Requirements Gathering Box structure Specification Formal Design Correctness Verification Code Generation, Inspect, Verify Statistical Test Planning Statistical Usage Testing Certification, Integration

32 How CoffeePot is this different (from other methods)? Statistical Q. C. Design Specs mathematically proved correct Statistical Usage Testing

33 Functional Specification: Box Structure: Referential transparency (information content stands alone) Black Box State Box Clear Box

34 Design Refinement: Structured Programming Stepwise expansion Verify at each level of refinement (Linger) Verification is a finite process Every line of design and code verified Very low defect levels It scales up Better code than unit testing Statistical Use Testing Probability Distribution Certification

35 Requirements: What to do. (what, NOT how.) Formal Methods: What is wrong with what is done now? Vague Ambiguity Incomplete Contradictions Mixed Abstraction and methods Clients must agree to requirements, Developers must be able to design from them. Word descriptions (suffer from above) Structured Analysis O-O Analysis

36 Use Mathematical Notation (Must extend some) Many choices available: CSP, VDM, Z, etc. Z ( Zed )

37 Should I Use Formal Methods? (Bowen and Hinchley) Use Appropriate Notation Formalize, not overformalize Costs? Have a FM guru available Don t abandon what you have Document Still need SQA No dogma Test, test, Reuse

38 What s good about Formal Methods: It is Real Engineering Precise Validation, Verification Testable Automated tools Good SE s can do it Don t need to spel gud or English Grammar Reusable (Many more)

39 What s not so good about Formal Methods: Problems?: Hard to learn I can t read this I can t write this Client can t either Is this the future? Lack of scoping and structuring

40 example Schema signature Schema predicate

41 CoffeePot Coffee: N PotCapacity: N Coffee PotCapacity

42 CoffeeLight Light: { off, on } CoffeeLevel: N PotCapacity: N Light = on Coffee > PotCapacity

43 CoffeeMaker CoffeePot FreshCoffee?: N FreshCoffee? + CoffeeLevel < PotCapacity CoffeeLevel = CoffeeLevel + FreshCoffee?

44 CoffeeSystem CoffeeMade CoffeeBrewing CoffeeOverflow

45 Some more notation: Domain to range: Coins = { cent 1, nickel 5 } I, j: dom s (I < j ) => s(i) < T s(j) ΞName means that state variables are unchanged

46 MakeChange CostOfItem: N ChangeFrom: N ChangeFrom = 1.00

47 Software Testing Object Oriented Testing

48 REMEMBER: What is the goal of testing? (Finding errors.) Assess quality Uncover errors Ensure standards are met

49 Is OO Test different? (From other methods) OO models are different Unit testing is based on classes Integration is different OO assumes re-use The basic goals of testing remain the same We want to find errors as early as possible (in process steps: requirements, design, code) to avoid propagation

50 Errors made in analysis may: Lead to errors in classes in design Make more work in procedures for a bad attribute (update, inherit, etc.) Cause extra (or incorrect) subclasses or system behavior How are analysis and design "models" tested Different methodologies (UML, Rumbaugh, Booch, etc) use different notation We still want to check that "some" correct syntax was used "Experts" in the problem domain can review class definitions and hierarchy for missing or ambiguous Do class relationships reflect "the real world"?

51 Class - Responsibility - Collaboration (CRC) model classes and relationships may be inconsistent CRC lists class name responsibilities (operations, what it does) collaborators (other classes needed, sent messages) and optionally: class type (like: transaction) class characteristics (systems related: atomic, secure)

52 CRC Steps: Cross check CRC and O-R model Ensure that a delegated responsibility is actually done by each (that) collaborator Ensure each collaborator receives requests from correct (reasonable) source Are other classes needed, is work grouped correctly? Can (should) widely requested services be grouped (combine responsibilities that always go together) NOTE: this may be a bad idea - reuse, etc.

53 OO Unit testing With encapsulation, smallest testable unit is a class (or object) How can we isolate and test? Class testing - unlike unit test based on algorithms and data flow - state behavior of class and encapsulated operations Integration testing What is top-down or bottom-up (with classes)? Integration often described as: Thread-based - based on response to an input or event. Use-based - test independent classes that use fewest server classes, then move on to dependant classes

54 Cluster testing collaborating classes are clustered (by O-R and CRC models) and exercised Validation testing At system level, class connections not visible Black Box works here Test cases may be done from OOA: Object Behavior, event flow

55 Test Case Design Berard Test case explicitly connected to class State test purpose Testing: Object states operations, messages exceptions tested May be difficult to know the internal state of an object It might be necessary to build in operations to report the state Inheritance requires retesting

56 Conventional testing (and differences): Fault based test - high probability of finding an error where are most likely errors (to be found)? Derived and Inherited classes - redefined class tests Scenario-based testing What does the user do? (Fault based - what does software do) Use Case Surface structure test - structure visible to user (commands, etc.) Deep Structure - examine design and code

57 Class Testing Random tests Partition test - similar to equivalence partition based on: Interclass test attribute class states category - functions of operations Multiple class testing random testing, partition testing Behavior model test based on state transition diagram (dynamic behavior) breadth first usually

58 Reasons not to test: You didn't know it was policy Blame someone else's code (design, etc.) Good choices: dead co-workers children that can't talk Bad choices: bosses children boss Not on the schedule We keep finding problems Too complex The risks (product liability) not great Waste of (my) time

59 REMEMBER: What is the goal of testing? (Finding errors.) Assess quality Uncover errors Ensure standards are met

60 Is OO Test different? (From other methods) OO models are different Unit testing is based on classes Integration is different OO assumes re-use The basic goals of testing remain the same We want to find errors as early as possible (in process steps: requirements, design, code) to avoid propagation

61 Errors made in analysis may: Lead to errors in classes in design Make more work in procedures for a bad attribute (update, inherit, etc.) Cause extra (or incorrect) subclasses or system behavior How are analysis and design "models" tested Different methodologies (UML, Rumbaugh, Booch, etc) use different notation We still want to check that "some" correct syntax was used "Experts" in the problem domain can review class definitions and hierarchy for missing or ambiguous Do class relationships reflect "the real world"?

62 Class - Responsibility - Collaboration (CRC) model classes and relationships may be inconsistent CRC lists class name responsibilities (operations, what it does) collaborators (other classes needed, sent messages) and optionally: class type (like: transaction) class characteristics (systems related: atomic, secure)

63 CRC Steps: Cross check CRC and O-R model Ensure that a delegated responsibility is actually done by each (that) collaborator Ensure each collaborator receives requests from correct (reasonable) source Are other classes needed, is work grouped correctly? Can (should) widely requested services be grouped (combine responsibilities that always go together) NOTE: this may be a bad idea - reuse, etc.

64 OO Unit testing With encapsulation, smallest testable unit is a class (or object) How can we isolate and test? Class testing - unlike unit test based on algorithms and data flow - state behavior of class and encapsulated operations Integration testing What is top-down or bottom-up (with classes)? Integration often described as: Thread-based - based on response to an input or event. Use-based - test independent classes that use fewest server classes, then move on to dependant classes

65 Cluster testing collaborating classes are clustered (by O-R and CRC models) and exercised Validation testing At system level, class connections not visible Black Box works here Test cases may be done from OOA: Object Behavior, event flow

66 Test Case Design Berard Test case explicitly connected to class State test purpose Testing: Object states operations, messages exceptions tested May be difficult to know the internal state of an object It might be necessary to build in operations to report the state Inheritance requires retesting

67 Conventional testing (and differences): Fault based test - high probability of finding an error where are most likely errors (to be found)? Derived and Inherited classes - redefined class tests Scenario-based testing What does the user do? (Fault based - what does software do) Use Case Surface structure test - structure visible to user (commands, etc.) Deep Structure - examine design and code

68 Class Testing Random tests Partition test - similar to equivalence partition based on: Interclass test attribute class states category - functions of operations Multiple class testing random testing, partition testing Behavior model test based on state transition diagram (dynamic behavior) breadth first usually

69 Reasons not to test: You didn't know it was policy Blame someone else's code (design, etc.) Good choices: dead co-workers children that can't talk Bad choices: bosses children boss Not on the schedule We keep finding problems Too complex The risks (product liability) not great Waste of (my) time

70

71 Re-engineering (This is a good "YAB" - Yet Another Buzzword) As the rules of business change the software supporting those rules must change. A Business Process ("BP") - like other processes is a set of related tasks to achieve a defined outcome BPR (Reengineering) is implementing change for "breakthroughs" BP has a "customer" - who receives the outcome (product)

72 Hammer suggests: Organize around outcomes (not tasks) (easier to "track" work status) The process customer performs (controls) the process. Put information processing into place where work is done. Get the data once, where it is produced. View distributed (geographically) resources as if they were centralized. Make decisions where work is done (not higher up).

73 BPR is a cycle, including: Define Business Goals Identify process, prioritize Evaluate processes, measure Specify and design process Prototype (try it out) Refine and use

74 Software Re-engineering: Maintenance is: Corrective Adaptive Perfective (enhance) Preventive (re-engineer) Process Model: Inventory Analysis - details of applications Document restructuring Reverse Engineer - create an abstract representation Restructure Code - most common Restructure Data - may actually be more valuable Forward Engineering - apply SE principles to recreate existing application (with new features)

75 Some details: Reverse Engineer to understand: Data Structures Database Model User IF Restructure to: Improve quality (how to measure?) Improve programmer productivity Maintainable, testable Forward Engineer to: Use new technology: Distributed, new GUI, OO methods, CASE Tools Is it worth it? Cost benefit is Maintenance costs - Re-engineering costs

76